decision making under support systems -...
TRANSCRIPT
709.049 08 Holzinger Group 1
Andreas HolzingerVO 709.049 Medical Informatics
14.12.2016 11:15‐12:45
Lecture 08 Decision Making under Uncertainty: Decision Support Systems
[email protected]: [email protected]
http://hci‐kdd.org/biomedical‐informatics‐big‐data
Science is to test crazy ideas – Engineering is to put these ideas into Business
709.049 08 Holzinger Group 2
Holzinger, A. 2014. Trends in Interactive Knowledge Discovery for Personalized Medicine: Cognitive Science meets Machine Learning. IEEE Intelligent Informatics Bulletin, 15, (1), 6‐14.
ML needs a concerted effort fostering integrated research
http://hci‐kdd.org/international‐expert‐network
709.049 08 Holzinger Group 3
Artificial intelligence Case based reasoning Computational methods in cancer detection Cybernetic approaches for diagnostics Decision support models Decision support system (DSS) Fuzzy sets MYCIN Reasoning under uncertainty Radiotherapy planning
Keywords
709.049 08 Holzinger Group 4
Case‐based reasoning (CBR) = process of solving new problems based on the solutions of similar past problems;
Certainty factor model (CF) = a method for managing uncertainty in rule‐based systems;
CLARION = Connectionist Learning with Adaptive Rule Induction ON‐line (CLARION) is a cognitive architecture that incorporates the distinction between implicit and explicit processes and focuses on capturing the interaction between these two types of processes. By focusing on this distinction, CLARION has been used to simulate several tasks in cognitive psychology and social psychology. CLARION has also been used to implement intelligent systems in artificial intelligence applications.
Clinical decision support (CDS) = process for enhancing health‐related decisions and actions with pertinent, organized clinical knowledge and patient information to improve health delivery;
Clinical Decision Support System (CDSS) = expert system that provides support to certain reasoning tasks, in the context of a clinical decision;
Collective Intelligence = shared group (symbolic) intelligence, emerging from cooperation/competition of many individuals, e.g. for consensus decision making;
Crowdsourcing = a combination of "crowd" and "outsourcing" coined by Jeff Howe (2006), and describes a distributed problem‐solving model; example for crowdsourcing is a public software beta‐test;
Advance Organizer (1/2)
709.049 08 Holzinger Group 5
Decision Making = central cognitive process in every medical activity, resulting in the selection of a final choice of action out of several alternatives;
Decision Support System (DSS) = is an IS including knowledge based systems to interactively support decision‐making activities, i.e. making data useful;
DXplain = a DSS from the Harvard Medical School, to assist making a diagnosis (clinical consultation), and also as an instructional instrument (education); provides a description of diseases, etiology, pathology, prognosis and up to 10 references for each disease;
Expert‐System = emulates the decision making processes of a human expert to solve complex problems;
GAMUTS in Radiology = Computer‐Supported list of common/uncommon differential diagnoses;
ILIAD = medical expert system, developed by the University of Utah, used as a teaching and testing tool for medical students in problem solving. Fields include Pediatrics, Internal Medicine, Oncology, Infectious Diseases, Gynecology, Pulmonology etc.
MYCIN = one of the early medical expert systems (Shortliffe (1970), Stanford) to identify bacteria causing severe infections, such as bacteremia and meningitis, and to recommend antibiotics, with the dosage adjusted for patient's body weight;
Reasoning = cognitive (thought) processes involved in making medical decisions (clinical reasoning, medical problem solving, diagnostic reasoning;
Advance Organizer (2/2)
709.049 08 Holzinger Group 6
… can apply your knowledge gained in the previous lectures to example systems of decision support;
… have an overview about the core principles and architecture of decision support systems;
… are familiar with the certainty factors as e.g. used in MYCIN;
… are aware of some design principles of DSS; … have seen similarities between DSS and KDD on the example of computational methods in cancer detection;
… have seen basics of CBR systems;
Learning Goals: At the end of this lecture you …
709.049 08 Holzinger Group 7
00 Reflection – follow‐up from last lecture 01 Decision Support Systems (DSS) 02 History of DSS = History of AI 03 Development of DSS 04 Further Practical Examples 05 Towards Precision Medicine (P4) 06 Case Based Reasoning (CBR)
Agenda for today
709.049 08 Holzinger Group 10
Fundamental Signal Detection Problem
10
Swets, J. A. 1961. Detection theory and psychophysics: a review. Psychometrika, 26, (1), 49‐63, doi:10.1007/BF02289684.
709.049 08 Holzinger Group 13
400 BC Hippocrates (460‐370 BC), father of western medicine: A medical record should accurately reflect the course of a disease A medical record should indicate the probable cause of a disease
1890 William Osler (1849‐1919), father of modern western medicine Medicine is a science of uncertainty and an art of probabilistic decision making
Today Prediction models are based on data features, patients are modelled as high‐dimensional feature vectors …
The Medical Domain
709.049 08 Holzinger Group 15
Our Example: Rheumatology
Chao, J., Parker, B. A. & Zvaifler, N. J. (2009) Accelerated Cutaneous Nodulosis Associated with Aromatase Inhibitor Therapy in a Patient with Rheumatoid Arthritis. The Journal of Rheumatology, 36, 5, 1087‐1088.
709.049 08 Holzinger Group 16
Bone Changes …
Ikari, K. & Momohara, S. (2005) Bone Changes in Rheumatoid Arthritis. New England Journal of Medicine, 353, 15, e13.
t
709.049 08 Holzinger Group 17
50+ Patients per day 5000 data points per day … Aggregated with specific scores (Disease Activity Score, DAS) Current patient status is related to previous data = convolution over time time‐series data
100+ clinical and functional parameter per Patient
Simonic, K. M., Holzinger, A., Bloice, M. & Hermann, J. (2011). Optimizing Long‐Term Treatment of Rheumatoid Arthritis with Systematic Documentation. Pervasive Health ‐ 5th International Conference on Pervasive Computing Technologies for Healthcare, Dublin, IEEE, 550‐554.
709.049 08 Holzinger Group 20
Medicine is an extremely complex application domain –dealing most of the time with probable information!
Some challenges include: (a) defining general system architectures in terms of generic tasks such as diagnosis, therapy planning and monitoring to be executed for (b) medical reasoning in (a);
(c) patient management with (d) minimum uncertainty. Other challenges include: (e) knowledge acquisition and encoding, (f) human‐computer interface and interaction; and (g) system integration into existing clinical environments, e.g. the enterprise hospital information system; to mention only a few.
Slide 8‐1 Key Challenges
709.049 08 Holzinger Group 21
Computers to help human doctors to make better decisions
http://biomedicalcomputationreview.org/content/clinical‐decision‐support‐providing‐quality‐healthcare‐help‐computer
709.049 08 Holzinger Group 22
Type 1 Decisions: related to the diagnosis, i.e. computers are used to assist in diagnosing a disease on the basis of the individual patient data. Questions include: What is the probability that this patient has a myocardial infarction
on the basis of given data (patient history, ECG, …)? What is the probability that this patient has acute appendices, given
the signs and symptoms concerning abdominal pain?
Type 2 Decisions: related to therapy, i.e. computers are used to select the best therapy on the basis of clinical evidence, e.g.: What is the best therapy for patients of age x and risks y, if an
obstruction of more than z % is seen in the left coronary artery? What amount of insulin should be prescribed for a patient during
the next 5 days, given the blood sugar levels and the amount of insulin taken during the recent weeks?
Slide 8‐2 Two types of decisions (Diagnosis vs. Therapy)
Bemmel, J. H. V. & Musen, M. A. 1997. Handbook of Medical Informatics, Heidelberg, Springer.
709.049 08 Holzinger Group 23
Example of a Decision Problem Soccer player considering knee surgery Uncertainties: Success: recovering full mobility Risks: infection in surgery (if so, needs another surgery and may loose
more mobility) Survival chances of surgery
Example: Knee Surgery of a Soccer Player
Harvard‐MIT Division of Health Sciences and Technology HST.951J: Medical Decision Support, Fall 2005Instructors: Professor Lucila Ohno‐Machado and Professor Staal Vinterbo
709.049 08 Holzinger Group 27
Clinical Decision Tree (CDT) is still state‐of‐the‐art
Ferrando, A., Pagano, E., Scaglione, L., Petrinco, M., Gregori, D. & Ciccone, G. (2009) A decision‐tree model to estimate the impact on cost‐effectiveness of a venous thromboembolism prophylaxis guideline. Quality and Safety in Health Care, 18, 4, 309‐313.
709.049 08 Holzinger Group 28
Slide 8‐3 Taxonomy of Decision Support Models
Bemmel, J. H. v. & Musen, M. A. (1997) Handbook of Medical Informatics. Heidelberg, Springer.
Decision Model
Quantitative (statistical) Qualitative (heuristic)
supervised
unsupervised
Neural network
Bayesian
Fuzzy sets
Logistic
Truth tables
Non‐parametricPartitioning
Decision trees
BooleanLogic
Reasoning models
Expert systems
Critiquing systems
709.049 08 Holzinger Group 30
Slide 8‐4 History of DSS is a history of artificial intelligence
Buchanan, B. G. & Feigenbaum, E. A. (1978) DENDRAL and META‐DENDRAL: their applications domain. Artificial Intelligence, 11, 1978, 5‐24.
E. Feigenbaum, J. Lederberg, B. Buchanan, E. Shortliffe
Rheingold, H. (1985) Tools for thought: the history and future of mind‐expanding technology. New York, Simon & Schuster.
709.049 08 Holzinger Group 31
Slide 8‐5 Evolution of Decision Support Systems
Shortliffe, E. H. & Buchanan, B. G. (1984) Rule‐based expert systems: the MYCIN experiments of the Stanford Heuristic Programming Project. Addison‐Wesley.
709.049 08 Holzinger Group 32
Slide 8‐6 Early Knowledge Based System Architecture
Shortliffe, T. & Davis, R. (1975) Some considerations for the implementation of knowledge‐based expert systems ACM SIGART Bulletin, 55, 9‐12.
709.049 08 Holzinger Group 33
Slide 8‐7 Static Knowledge versus dynamic knowledge
Shortliffe & Buchanan (1984)
709.049 08 Holzinger Group 34
The information available to humans is often imperfect – imprecise ‐ uncertain. This is especially in the medical domain the case. An human agent can cope with deficiencies. Classical logic permits only exact reasoning: IF A is true THEN A is non‐false and IF B is false THEN B is non‐true Most real‐world problems do not provide this exact information, mostly it is inexact, incomplete, uncertain and/or un‐measurable!
Slide 8‐8 Dealing with uncertainty in the real world
709.049 08 Holzinger Group 36
MYCIN is a rule‐based Expert System, which is used for therapy planning for patients with bacterial infections
Goal oriented strategy (“Rückwärtsverkettung”) To every rule and every entry a certainty factor (CF) is assigned, which is between 0 und 1
Two measures are derived: MB: measure of belief MD: measure of disbelief Certainty factor – CF of an element is calculated by:
CF[h] = MB[h] – MD[h] CF is positive, if more evidence is given for a hypothesis, otherwise CF is negative
CF[h] = +1 ‐> h is 100 % true CF[h] = –1 ‐> h is 100% false
Slide 8‐9 MYCIN – rule based system ‐ certainty factors
709.049 08 Holzinger Group 37
Slide 8‐10 Original Example from MYCIN
Shortliffe, E. H. & Buchanan, B. G. (1984) Rule‐based expert systems: the MYCIN experiments of the Stanford Heuristic Programming Project. Addison‐Wesley.
709.049 08 Holzinger Group 40
Slide 8‐12 Basic Design Principles of a DSS
Majumder, D. D. & Bhattacharya, M. (2000) Cybernetic approach to medical technology: application to cancer screening and other diagnostics. Kybernetes, 29, 7/8, 871‐895.
709.049 08 Holzinger Group 41
Slide 8‐13 Cybernetic approach to medical diagnostics
Majumder, D. D. & Bhattacharya, M. (2000) Cybernetic approach to medical technology: application to cancer screening and other diagnostics. Kybernetes, 29, 7/8, 871‐895.
709.049 08 Holzinger Group 42
Slide 8‐14 State‐of‐the‐art architecture of DSS
Metaxiotis, K. & Psarras, J. (2003) Expert systems in business: applications and future directions for the operations researcher. Industrial Management & Data Systems, 103, 5, 361‐368.
709.049 08 Holzinger Group 43
Human–Computer cooperation (integration!) is essential to the decision support process.
Consequently, the field of Human–Computer Interaction (HCI) is a fundamental aspect for building
intelligent, interactive DSS, because the design of such systems heavily relies on a user‐centered approach (usability of machine learning!).
It is necessary to combine and integrate methods from Software Engineering (SE) and HCI.
Traditional methods and models are limited because the system is highly interactive and
usually these methods do not integrate the end‐user explicitly and systematically.
Slide 8‐15 On design and development of DSS
709.049 08 Holzinger Group 45
Slide 8‐16 Example: Development following the U model 1/2
Abed, M., Bernard, J. & Angué, J. (1991). Task analysis and modelization by using SADT and Petri Networks. Tenth European Annual Conference on Human Decision Making and Manual Control, Liege, 11‐13.
Unified Process (UP) from SE and the U‐model from HCI.
709.049 08 Holzinger Group 46
Slide 8‐17 Improved U model 2/2
Ayed, B. M., Ltifi, H., Kolski, C. & Alimi, A. (2010) A user‐centered approach for the design & implementation of KDD‐based DSS: A case study in the healthcare domain. Decision Support Systems, 50, 64‐78.
709.049 08 Holzinger Group 47
Slide 8‐18 Remember the similarities between DSS and KDD
Ayed et al. (2010)
709.049 08 Holzinger Group 51
Clinical guidelines are systematically developed documents to assist doctors and patient decisions about appropriate care;
In order to build DS, based on a guideline, it is formalized(transformed from natural language to a logical algorithm), and
implemented (using the algorithm to program a DSS); To increase the quality of care, they must be linked to a process
of care, for example: “80% of diabetic patients should have an HbA1c below 7.0” could be
linked to processes such as: “All diabetic patients should have an annual HbA1c test” and “Patients with values over 7.0 should be rechecked within 2 months.”
Condition‐action rules specify one or a few conditions which are linked to a specific action, in contrast to narrative guidelines which describe a series of branching or iterative decisions unfolding over time.
Narrative guidelines and clinical rules are two ends of a continuum of clinical care standards.
Slide 8‐20 Clinical Guidelines as DSS & Quality Measure
709.049 08 Holzinger Group 52
Slide 8‐21 Clinical Guidelines
Medlock, S., Opondo, D., Eslami, S., Askari, M., Wierenga, P., de Rooij, S. E. & Abu‐Hanna, A. (2011) LERM (Logical Elements Rule Method): A method for assessing and formalizing clinical rules for decision support. International Journal of Medical Informatics, 80, 4, 286‐295.
709.049 08 Holzinger Group 53
Slide 8‐21b Gamuts: Triangulation to find diagnoses
Reeder, M. M. & Felson, B. 2003. Reeder and Felson's gamuts in radiology: comprehensive lists of roentgen differential diagnosis, New York, Springer Verlag.
709.049 08 Holzinger Group 54
Slide ‐21c Example ‐ Gamuts in Radiology
http://rfs.acr.org/gamuts/data/G‐25.htm
Reeder, M. M. & Felson, B. (2003) Reeder and Felson's gamuts in radiology: comprehensive lists of roentgen differential diagnosis. New York, Springer Verlag.
709.049 08 Holzinger Group 56
Slide 8‐22 Example: Exon Arrays
Kapur, K., Xing, Y., Ouyang, Z. & Wong, W. (2007) Exon arrays provide accurate assessments of gene expression. Genome Biology, 8, 5, R82.
709.049 08 Holzinger Group 57
Slide 8‐23 Computational leukemia cancer detection 1/6
Exon array structure. Probe design of exon arrays. (1) Exon—intron structure of a gene. Gray boxes represent introns, rest represent exons. Introns are not drawn to scale. (2) Probe design of exon arrays. Four probes target each putative exon. (3) Probe design of 30expression arrays. Probe target the 30end of mRNA sequence.
Corchado, J. M., De Paz, J. F., Rodriguez, S. & Bajo, J. (2009) Model of experts for decision support in the diagnosis of leukemia patients. Artificial Intelligence in Medicine, 46, 3, 179‐200.
709.049 08 Holzinger Group 58
Slide 8‐24 Computational leukemia cancer detection 2/6
Corchado et al. (2009)
709.049 08 Holzinger Group 59
Slide 8‐25 Computational leukemia cancer detection 3/6
Corchado et al. (2009)
• ALL = cancer of the blood AND bone marrow caused by an abnormal proliferation of lymphocytes.
• AML = cancer in the bone marrow characterized by the proliferation of myeloblasts, red blood cells or abnormal platelets.
• CLL = cancer characterized by a proliferation of lymphocytes in the bone marrow.
• CML = caused by a proliferation of white blood cells in the bone marrow.
• MDS (Myelodysplastic Syndromes) = a group of diseases of the blood and bone marrow in which the bone marrow does not produce a sufficient amount of healthy cells.
• NOL (Normal) = No leukemias
A = acute, C = chronic, L = lymphocytic, M = myeloid
709.049 08 Holzinger Group 60
8‐26 Computational leukemia cancer detection 4/6
Corchado et al. (2009)
Further Reading: Breiman, Friedman, Olshen, & Stone (1984). Classification and Regression Trees. Wadsworth, Belmont, CA.
709.049 08 Holzinger Group 61
8‐27 Computational leukemia cancer detection 5/6
Classification CLL—ALL. Representation of the probes of the decision tree which classify the CLL and ALL to 1555158_at, 1553279_at and 1552334_at
Corchado et al. (2009)
709.049 08 Holzinger Group 62
The model of Corchado et al. (2009) combines: 1) methods to reduce the dimensionality of the original data set;
2) pre‐processing and data filtering techniques; 3) a clustering method to classify patients; and 4) extraction of knowledge techniques The system reflects how human experts work in a lab, but
1) reduces the time for making predictions; 2) reduces the rate of human error; and 3) works with high‐dimensional data from exon arrays
Slide 8‐28 Computational leukemia cancer detection 6/6
709.049 08 Holzinger Group 65
Slide 8‐30 Case Based Reasoning (CBR) Basic principle
Aamodt, A. & Plaza, E. (1994) Case‐based reasoning: Foundational issues, methodological variations, and system approaches. AI Communications, 7, 1, 39‐59.
709.049 08 Holzinger Group 67
Slide 8‐32 CBR Example: Radiotherapy Planning 1/6
Source: http://www.teachingmedicalphysics.org.uk
709.049 08 Holzinger Group 68
Slide 8‐33 CBR Example: Radiotherapy Planning 2/6
Source: Imaging Performance Assessment of CT Scanners Group, http://www.impactscan.org
709.049 08 Holzinger Group 69
Slide 8‐34 CBR Example: Radiotherapy Planning 3/6
Petrovic, S., Mishra, N. & Sundar, S. (2011) A novel case based reasoning approach to radiotherapy planning. Expert Systems With Applications, 38, 9, 10759‐10769.
Measures:1) Clinical Stage = a labelling system2) Gleason Score = grade of prostate cancer = integer between 1 to 10; and 3) Prostate Specific Antigen (PSA) value between 1 to 404) Dose Volume Histogram (DVH) = pot. risk to the rectum (66, 50, 25, 10 %)
709.049 08 Holzinger Group 70
Slide 8‐35 CBR System Architecture 4/6
Petrovic, S., Mishra, N. & Sundar, S. (2011) A novel case based reasoning approach to radiotherapy planning. Expert Systems With Applications, 38, 9, 10759‐10769.
709.049 08 Holzinger Group 71
Slide 8‐36 Membership funct. of fuzzy sets Gleason score 5/6
Petrovic, S., Mishra, N. & Sundar, S. (2011) A novel case based reasoning approach to radiotherapy planning. Expert Systems With Applications, 38, 9, 10759‐10769.
Gleason score evaluates the grade of prostate cancer. Values: integer within the range
709.049 08 Holzinger Group 75
Again: What is human intelligence? What different decision models can be applied in medical informatics?
How can we deal with uncertainty in the real world? What is the principle of a rule based expert system? Sketch the basic architecture of a DSS! Which basic design principles of a DSS must be considered?
How does the U‐model work to engineer an DSS? Which similarities exist between DSS and KDD? What are clinical guidelines? What is interesting in the computational method model of cancer detection of Corchado et al. (2009)?
What is the basic principle of Case Based Reasoning?
Sample Questions
709.049 08 Holzinger Group 76
Why was MYCIN no success in the practical clinical world? What are clinical guidelines? What is a big advantage of the Gamuts‐for‐Radiology System? Why are decision trees so important for DSS? Why would radiotherapy planning without computers so difficult?
Sample Questions
709.049 08 Holzinger Group 78
http://gaia.fdi.ucm.es/projects/jcolibri http://www‐formal.stanford.edu/jmc/whatisai/whatisai.html http://aaai.org/AITopics http://www.stottlerhenke.com/ai_general/history.htm http://rfs.acr.org/gamuts (Gamuts in radiology ‐ DSS for
radiological imaging) http://www.scribd.com/doc/16093558/Gamuts‐in‐radiology
(Reeder & Felsons Original Book on Gamuts) http://www.isradiology.org/gamuts/Gamuts.htm (Web‐based
Gamuts in Radiology)
Some Useful Links
709.049 08 Holzinger Group 80
Appendix: CDS Tools and EHR for quality measures
Downing, G., Boyle, S., Brinner, K. & Osheroff, J. (2009) Information management to enable personalized medicine: stakeholder roles in building clinical decision support. BMC Medical Informatics and Decision Making, 9, 1, 1‐11.
709.049 08 Holzinger Group 81
Appendix: Quality Improvement and Health Records
Yu, F.B., Allison, J.J., Houston T.K. (2008) Quality Improvement & the Electronic Health Record: Concepts and Methods. In: Carter, J. H. (2008) Electronic health records: a guide for clinicians & administrators. ACP Press.
709.049 08 Holzinger Group 82
Appendix: Example for a Knowledge discovery module
Funk, P. & Xiong, N. (2006) Case‐based reasoning and knowledge discovery in medical applications with time series. Computational Intelligence, 22, 3‐4, 238‐253.
709.049 08 Holzinger Group 83
Appendix: Discovery of new expressions & co‐occurrences
Funk & Xiong (2006)
709.049 08 Holzinger Group 84
Appendix: Evaluation of a Sequence
Funk & Xiong (2006)
Given a sequence s there may be a set of probable consequent classes {C1, C2, . . . ,Ck}
The strength of the co‐occurrence between sequence s and class Ci (i = 1, . . . , k) can be measured by the probability, p(Ci | s), of Ci conditioned upon s
709.049 08 Holzinger Group 85
Appendix: Example for Interactive Group Decision Support
Anya, O., Tawfik, H. & Nagar, A. (2011) Cross‐boundary knowledge‐based decision support in e‐health. International Conference on Innovations in Information Technology (IIT). 150‐155.
709.049 08 Holzinger Group 86
Appendix: Crowdsourcing ‐ Example Disaster Management
Gao, H., Barbier, G. & Goolsby, R. (2011) Harnessing the Crowdsourcing Power of Social Media for Disaster Relief. Intelligent Systems, IEEE, 26, 3, 10‐14.
709.049 08 Holzinger Group 89
Example (Part B): Leukemia
This 79 y/o female with chronic myeloid leukemia presented with rapidly enlarging spleen. The splenectomy specimen showed a dark red surface devoid of white pulp. Majority of the large tumor cells seen here were positive for CD34. This is a case of chronic myeloid leukemia in blast transformation (Richter’s Syndrome) Source: webpathology.com
709.049 08 Holzinger Group 90
Gamuts: Triangulation to find diagnoses
Reeder, M. M. & Felson, B. 2003. Reeder and Felson's gamuts in radiology: comprehensive lists of roentgen differential diagnosis, New York, Springer Verlag.
709.049 08 Holzinger Group 91
Example ‐ Gamuts in Radiology
http://rfs.acr.org/gamuts/data/G‐25.htm
Reeder, M. M. & Felson, B. (2003) Reeder and Felson's gamuts in radiology: comprehensive lists of roentgen differential diagnosis. New York, Springer Verlag.
709.049 08 Holzinger Group 92
Type 1: Decisions related to the diagnosis, i.e. computers are used to assist in diagnosing a disease on the basis of the individual patient data. Questions include: What is the probability that this patient has a myocardial infarction on the basis of given data (patient history, ECG)? What is the probability that this patient has acute appendices, given the signs and symptoms concerning abdominal pain?
Two types of decisions (Diagnosis vs. Therapy)
Bemmel, J. H. V. & Musen, M. A. 1997. Handbook of Medical Informatics, Heidelberg, Springer.
709.049 08 Holzinger Group 93
Type 2: Decisions related to therapy, i.e. computers are used to select the best therapy on the basis of clinical evidence, e.g.: What is the best therapy for patients of age x and risks y, if an obstruction of more than z % is seen in the left coronary artery? What amount of insulin should be prescribed for a patient during the next 5 days, given the blood sugar levels and the amount of insulin taken during the recent weeks?
Two types of decisions (Diagnosis vs. Therapy)
709.049 08 Holzinger Group 94
Sometimes we do not have “big data”, where aML‐algorithms benefit. Sometimes we have Small number of data sets Rare events NP‐hard problems (e.g. k‐Anonymization, Protein‐Folding, Graph Coloring, Subspace Clustering, …)
Then we still need the “human‐in‐the‐loop”
Slide 8‐39 Future Outlook